Itay M wrote:
I'm a bit confused by the new nifty settings that you've just
suggested, and by the MAUI/TORQUE relationship with it. Maybe I should
expand my question:

We have one MAUI/Torque server with 3 queues. The user groups have a
MAXPROC soft/hard values (according to our usage policy). Is the
following configuration suggested?  -

In qmgr :
* unset the resources_max.ncpus from both server and queue levels

I don't think resources_max.ncpus really matters, but I do suggest unsetting it "just in case" for the sake of troubleshooting your problem. I guess that you can set it back afterwards to enforce your policy.

* unset the resources_default.ncpus from both server and queue levels

Yes.

* should I make sure any other settings are unset?

Nothing that I can think of right now. Basically, what we are doing is troubleshooting by making your configuration closer to a known working one (mine). So if the recommended settings do not work, we can compare some more.

In qsub:
* Ask the users that needs 4 processors on the same node to use the
nodes=1:ppn=4 (instead of "NCPUS=4)

Yes.

Question is:
If I remove the resources_max.ncpus from the queue level - how will
MAUI/PBS know how much procs does the entire cluster has? Should I set
it somewhere else not mentioned here?

In the server_priv directory of the TORQUE installation there is a file named 'nodes' in which the names and features of all nodes should be listed. Here are two example lines from our cluster:

node2 np=8 ib quad prod
node3 np=4 ib dual prod

The np parameter specifies the number of processors. "ib", "quad" and "prod" are attributes to which you can refer from your resource requirements, for example:

-l nodes=1:ib:ppn=4

to request 4 processors on a node with the attribute "ib" (the attribute's meaning is "InfiniBand-equipped" in our cluster).

Regards,
Jan Ploski
_______________________________________________
mauiusers mailing list
mauiusers@supercluster.org
http://www.supercluster.org/mailman/listinfo/mauiusers

Reply via email to